INTERPRET NTA User Guide

Data Types and Quality

Data input to INTERPRET NTA is expected to be from LC-HRMS experiments, where chemical features are measured as an accurate mass (e.g., 413.9737 g/mol) at a specific retention time (e.g., 6.2 minutes). Data acquired from low-resolution or other mass spectrometry platforms is not suitable for INTERPET NTA processing.

The MS1 Data Processing workflow requires only MS1-level data (e.g., feature lists). This means HRMS data that can be input into INTERPRET NTA may be from MS1-only (e.g., no fragmentation) data acquisition or may have acquired MS2 data. Studies with MS2 data can have data acquired through either data-dependant (DDA) and data-independent (DIA) acquisition methods.

INTERPRET NTA uses both mass tolerance and retention time tolerance windows to aide in identification of tracers, adducts, neutral losses, and duplicate features. To ensure these tolerances are utilized properly, input data must have the appropriate precision relative to the measurement technique used, here LC-HRMS. Mass tolerances can be set in units of Da or ppm and should be set with an understanding of the typical performance of the instrument and methods utilized. For most HRMS systems this will typically be no more than 0.002 Da or 10 ppm but may be smaller for select HRMS platforms or instruments.

Retention time tolerances are set with units of minutes and should be chosen based on an understanding of the typical performance of the instrument and methods utilized. Most reverse-phase liquid chromatography methods will produce peak widths in the range of 5-20 seconds with the variability in observed retention time less than 10-15 seconds between injections.Isomeric chemicals may produce many peaks across a longer time (30 seconds or more). Users should consider the potential presence of isomeric chemicals in their data and how they wish to treat (e.g., consider together or separately) these features in their data processing.

Spiked Compounds (Tracers)

INTERPRET NTA utilizes known compounds spiked into samples, termed tracers, to assist in data quality review. Tracer chemicals can take a variety of forms, primarily either as native compounds spiked into select control samples (e.g., surrogate chemicals, system suitability mixtures, etc…) or as internal standards spiked into all sample extracts. The use of these different types of spiked compounds as tracer chemicals is described in more detail below.

Native Compounds: are chemicals that are spiked into select control samples (e.g., matrix spikes, method spikes, laboratory controls, etc…) and represent at least some of the chemicals of interest for the NTA study. These chemicals could be found in study samples and are often included in order to determine chromatographic, MS1, MS2, and method performance parameters for chemicals of interest.

Internal Standards: are chemical standards that are spiked into samples (before and/or after extraction) that can be used to identify analyte loss or gain during sample preparation, define matrix effects, and ultimately enable accurate quantification of native compounds. Internal standards should not be present naturally in samples and should produce a stable, validated response when using the method. Internal standards should properly represent the chemical space of your analytes of interest and have applicability to the native compounds included in the study. Isotopically-labeled analogs of representative compounds fulfill the role of internal standards well and offer the best representation of matrix effects on study analytes, while not being found in study samples. Unfortunately, not all chemicals have isotopically-labeled internal standards available for purchase.

Types of Blanks

Blanks are critical to an NTA study to ensure chemicals identified in samples are attributable to to study samples and are not contaminants introduced during the collection, preparation, or analysis of samples. Many studies will utilize a variety of blank types (e.g., solvent, trip, field, equipment, lab, method, or matrix) to assess and measure different sources or routes of contamination. Table 1 below, adapted from Whitehead et al. (2025) defines different types of blanks that are commonly employed in an NTA study.

Table 1. Different blanks that can be utilized in a typical NTA study.

Name	Definition	Purpose
Trip Blank	Closed container of laboratory-grade water (or a matched blank matrix) that travels to the field location(s), remaining closed, that undergoes extraction & preparation alongside samples	Identifies contaminants introduced from sampling containers
Field Blank	Container of laboratory-grade water (or a matched blank matrix) that travels to the field location(s), is opened during sampling, that undergoes extraction & preparation alongside samples	Identifies contaminants introduced during the act of sampling
Equipment Blank	Container of laboratory-grade water (or a matched blank matrix) that is processed through the same equipment used during sampling, that undergoes extraction & preparation alongside samples	Identifies contaminants introduced by specific tools or steps in sampling (e.g., pumps)
Method Blank	Laboratory-grade water (or a matched blank matrix) that undergoes extraction & preparation alongside samples, spiked with internal standards used in the study	Identifies contaminants introduced during sample preparation and measures the extraction efficiency of the sample preparation method without matrix
Laboratory Blank	Diluent used to prepare final sample extracts, spiked with internal standards used in the study	Measures the response of isotopically-labeled compounds in the same solvent composition as samples (excluding matrix)
Solvent Blank	Diluent used to prepare final sample extracts, not spiked with any native or internal standards used in the study	Identifies contaminants present in the diluent or the instrument, due to carryover or contamination of mobile phases

Blanks are also critical to the functionality of QA/QC filtering performed by INTERPRET NTA where blanks are used to perform (1) MRL calculations and (2) blank subtraction.

A minimum reporting limit (MRL) can be defined as the smallest amount of a substance that can be reliably measured using a given analytical method. Calculation of an MRL is an important step for studies reporting both the identification and/or the quantification of chemicals in a sample. The MRL calculation performed in INTERPRET NTA is described in more detail in the INTERPRET NTA Logic section below, but briefly is determined for a given feature using the measured abundances in blank replicates to calculate the mean blank abundance (\(\widehat{μ}\)). The user then selects an MRL standard deviation (\(\widehat{σ}\)) multiplier (M) of 3, 5, or 10. The MRL for each feature is then calculated as the mean blank abundance plus multiplier times standard deviation, as shown below.

MRL=\(\widehat{μ}\)+(M×\(\widehat{σ}\))

Assuming that blank detection abundances for any feature are normally distributed, M values of 3, 5, and 10 allow approximation of the upper (one-tailed) 97.5th, 99.5th, and 99.9th percentiles of the blank abundance distribution. A MRL is calculated for all features where a reproducible abundance is measured in blank replicates. Features where no signal is observed in the blanks have an MRL set to 0. The measured abundance of features in samples is compared to the calculated MRL to ensure that abundances found in samples are larger than those in blank replicates.

In addition to MRL checks, INTERPRET NTA uses blank subtraction to remove feature abundances observed in blank replicates from study samples. While blank subtraction is commonly used in traditional quantitative analyses (e.g., targeted LC-MS/MS), its use and application in NTA experiments is typically dependent on the individual study and researcher. In INTERPRET NTA blank subtraction serves to remove the contribution of background signal of features from samples occurrences that have passed MRL checks. Blank subtraction is described in more detail in the INTERPRET NTA Logic section below, but briefly is determined by subtracting the mean blank abundance (\(\widehat{μ}\)_blank) from the mean sample abundance (\(\widehat{μ}\)_sample) for a given sample replicate.

Blank subtracted sample abundance= \(\widehat{μ}\)_sample - \(\widehat{μ}\)_blank

The blank(s) used are decided by the user and should represent the most complete and relevant background possible. For example, a laboratory blank made with just the final extract diluent will contain background features introduced from the solvent(s) used to prepare the diluent. In comparison, a method blank containing a matched “blank” matrix that undergoes the same preparation steps as the samples is expected to contain any background features introduced during sample preparation, including those from the solvent(s) used to prepare the diluent. In this example using a method blank would more appropriately capture all potential background features than the laboratory blank.

When isotopically-labeled compounds are utilized in a study, the blank(s) used for processing should also contain these compounds at a spiking concentration identical to that in study samples to allow direct assessment of method performance with INTERPRET NTA’s tracer functionality. For more information on the tracer functionality, please see the Tracers section below.

Studies with more than one blank type should decide which blank(s) should be used to perform MRL checks and blank subtraction. The blank(s) chosen should be identified using specific character strings in their sample names, as described in the Detection Matrices section below. Additional blanks that are not chosen may be kept in the input data and treated as samples for INTERPRET NTA processing. Ensure any blank(s) not chosen for MRL checks and blank subtraction have names that do not include the specific character strings described in the Detection Matrices section below.

Sample and Blank Replicates

The QA/QC filtering logic described in the INTERPRET NTA Logic below relies on the use of both blank and sample replicates to perform replicate threshold, coefficient of variation (CV), and MRL filtering, as well as blank subtraction. Replicate data may arise from the collection of multiple samples from a single source during sample collection (sample replicates), the preparation of multiple extracts from a single sample during sample preparation (preparation replicates), or, from repeated injection and acquisition of a single extract during data acquisition (analytical replicates or injection replicates).

INTERPRET NTA users must include, at minimum, three analytical replicates per sample and blank. If replicate analysis occurs across multiple analytical batches, users are encouraged to include replicates across batches and to randomize replicates within each batch. The use of analytical batches and the randomization of samples is described in more detail in the ENTAiLS Toolkit.

Summary
An analytical batch can be defined as a set of study and QA/QC samples analyzed together on an instrument. Small studies may have few enough samples that all study and QA/QC samples can be analyzed within a single analytical batch. However, it is common for studies to require multiple analytical batches spread across days, weeks, or months, and the use of analytical batches can ensure that instrument/signal drift is minimized, giving sufficient “down time’ where calibration and routine maintenance can be performed. Instrument performance can be tracked between batches using QA/QC samples and elements (e.g., spiked standards, controls). Randomizing the order of which samples are injected within and across these batches can be helpful to minimize performance effects for entire groups of similar study or QA/QC samples.

Sample replicates must be identified accordingly in the input detection matrix to ensure correct grouping and processing with INTERPRET NTA. Sample and blank replicates should be identified by separation of a single character, for example, ‘My_Sample_a’ and ‘My_Sample_b’, or “Blank_a” and “Blank_b”. This may be done using letters (A-Z) or with numbers (0-9).

Detection Matrices

The primary input to the MS1 workflow is a detection matrix containing a list of MS1 features identified using an upstream data processing software. This software may be vendor specific (e.g., Compound Discoverer, Mass Profiler, MassHunter, Progenesis QI, or SCIEX OS) or open source (e.g., mzmine, MS-DIAL, XCMS). The detection matrix generated should be a N by M matrix, where N represents the total number of features identified (as rows) and M represents the total number of samples (as columns). Each row should be for a unique feature (e.g., a unique mass and retention time pair). In addition to the sample-level data (e.g., observed peak areas), columns should also contain feature-level data (mass or m/z, retention time, unique identifiers). These required columns are summarized in Table 2. An example negative mode detection matrix is attached here.

Download detectionMatrix.csv

Items to note:

1. Feature-level information columns are typically found in the first few columns, though an exact order is not required
   
2. Peak areas should not be imputed, normalized, or otherwise transformed prior to processing to ensure efficacy of the replicate, CV, and MRL filtering steps
   
3. Non-detects (e.g., sample occurrences where a feature was not found by the upstream data processing software) can be input as empty cells, cells with 0 input as the peak area, or cells with another constant value input
   
4. The user must identify how non-detects are input in the input parameters, as described in the Input Parameters section below

Additional columns (e.g., initial annotations from upstream data processing, other metadata) beyond those described above may also be included in the input detection matrix and will be treated as pass-through columns that are unaltered during processing.

Table 2.Required columns and their description in the input Positive and Negative mode files (detection matrices).

Required column	Acceptable column names	Description of column contents	Additional comments
Mass	Mass, mass, MASS	The neutral, monoisotopic mass of a feature. Submitted values should have an appropriate number of decimal places relative to the measurement technique (e.g., at least four for HRMS data)	This mass should be reflective of any upstream data processing software or tools that were used to identify adducts/multimers/other ion forms
m/z	m/z	The measured mass-to-charge (m/z) value of a feature. This value should have an appropriate number of decimal places relative to the measurement technique (e.g., at least four for HRMS data)	If a “Mass” column is present, the m/z value will be treated as a pass-through column. If mass is not present this column will be used by INTERPRET NTA (along with the ionization mode) to calculate the feature’s mass. The m/z value will not be reflective of any already identified adducts/multimers/other ion forms from upstream processing
Retention_Time	Retention_time, Retention_Time, RETENTION_TIME, rt, RT, Ret._Time	The measured retention time of a feature. This value should be given in units of minutes and have an appropriate number of decimal places relative to the measurement technique (e.g., one or two for LC data)
Sample or Blank abundance	For blanks used for MRL determination and blank subtraction: MB, mb, mB, Mb, BLANK, Blank, blank	The measured response of features in each sample or bank, typically given as an integrated peak area	If any blank (e.g., solvent blank) is not desired to used be for MRL determination and blank subtraction, ensure none of the “acceptable column names” are used. Sample responses should not be imputed, normalized, or otherwise transformed prior to processing

Run Sequence

The run sequence file(s) contain a list of all samples included in the input detection matrices alongside sample groupings that identify the samples within user-defined groups. The requirements of this list includes the name of samples (exactly matching the names used in the input detection matrices) and the sample grouping as summarized in Table 3. The groups included in the run sequence file will be used to group samples on the interactive and static run sequence plots generated by INTERPRET NTA.

When both positive and negative detection matrices are input a run sequence file should be prepared for both modes. An example run sequence file that is paired to the example input detection matrix is attached below.

Download runSequence.csv

Table 3.Required columns and their description in the input Run Sequence files.

Required column	Acceptable column names	Description of column contents	Additional comments
Run sequence	Sample Sequence	Each row contains a different sample name that exactly matches a sample name in the relevant input detection matrix	The run sequence order is derived from the order in which samples are listed (e.g., row 2 is the first sample injected, row 3 the second, and so on)
Grouping	Sample Group	A user-defined sample grouping to identify samples within a group. Cells should contain a string that describes the sample type or other group metric (e.g., Blank, Sample, Control)	Maximum of 10 unique groups

Tracers

The tracer file contains a list of tracer compounds spiked into some or all samples and blanks at detectable concentrations. Requirements of the tracer file includes the name, anticipated ionization mode, exact mass, and the expected retention time of each tracer (Table 4). The anticipated ionization mode column is used to designate is a tracer chemical is expected in positive or negative mode. If a tracer chemical is expected to ionize in both modes, the input tracer file should have two lines, one with ESI+ and one with ESI- used in the anticipated ionization mode column.

Download tracerFile.csv

Table 4.Required columns and their description in the input Tracer files.

Required column	Acceptable column names	Description of column contents	Additional comments
Name	Chemical_Name	An identifier given for a chemical. Can be an IUPAC name, general name, acronym, or other unique identifier
Ionization mode	Ionization_Mode	The ionization mode for which a tracer is expected to be observed as a feature. Should be given as either “ESI-” or “ESI+”	When a tracer is amenable to both ESI+ and ESI-, and data from both modes are included as input detection matrices, that tracer should present twice in the outputs files, with one entry per ionization mode
Mass	Monoisotopic_Mass	The exact mass of a tracer. This value should have an appropriate number of decimal places relative to the measurement technique (e.g., at least four for HRMS data)	Ensure compatibility between mass provided in input detection matrices and tracer files. Certain tracers may be best monitored by searching for adducts or alternative ion forms instead of the addition or subtraction of a proton (e.g., GenX is a PFAS that is typically monitored as [M-H-CO2]-). If adducts and other ion forms have not been accounted for in the input detection matrices, the tracer exact mass values may need to be adjusted accordingly
Retention time	Retention_Time	The measured retention time of a tracer. This value should be given in units of minutes and have an appropriate number of decimal places relative to the measurement technique (e.g., one or two for LC data)	The retention times given here should be reflective of the expected retention time based on previous method experience or validation

Depending on experimental design and study objectives, separate tracer input files may be prepared for spiked native and internal standards. If separate tracer files are prepared, the user should run INTERPRET NTA twice, once with each tracer file, to separately capture visual and tabular results.

Additional columns beyond those described may also be included in the tracer file. These columns will be treated as pass-through columns with their contents unaltered during INTERPRET NTA processing. An example tracer file is attached below. This example matrix contains the isotopically-labeled internal standards present in EPA Method 1633. One additional column, “Formula”, is included as a pass-through column to track the DSSTox Substance Identifier (DTXSID) of each compound through steps of INTERPRET NTA processing.

Data Structure Checks

Once the input files and parameters are submitted, the first thing that happens is checking for the required columns, normalizing column names, and sorting columns alphabetically. In greater detail, the application first normalizes all column names by coercing all column names to strings, stripping leading and trailing spaces, and removing non-breaking spaces.

The application then verifies that an ionization mode column is present (checks for “Ionization_Mode”; if not present, assign “Ionization_Mode” based on what selector was used to upload the file), a mass column is present (either “Mass”, or “m/z” that is then converted to “Mass” based on the ionization mode), and a retention time column is present (will rename any of “Retention_time”, “RETENTION_TIME”, “Retention time”, “Retention Time”, “rt”, “RT”, and “Ret._Time” to “Retention_Time”). If a “Mass” or “m/z” column is not present, the run will stop; if any of the accepted derivatives of “Retention_Time” are not present, the run will stop. Following the column checks, the columns are sorted alphabetically such that all sample replicates should be adjacent to one another if the correct naming convention was followed.

After formatting and column verification, the input data is assigned a new column, “Feature ID”, which provides each chemical feature (i.e., row of data) a unique numerical value. The “Feature ID” column is used throughout the application to track data, perform merges, and reassign associated metadata that may have been included in the input files (see below).

The next step is parsing the column names in the input files to group together sample replicates and identify blanks. Blanks are identified by containing any of the following strings: “Blank”, “blank”, “BLANK”, “MB”, “Mb”, “mb”, “mB” (NOTE: if a sample name contains any of these strings, it can be accidentally identified as a blank). Currently, the application logically separates blank and sample columns from other metadata by asserting that the groups are greater than 1 – this excludes metadata columns but is a soft requirement that samples be replicated in order to successfully be processed by INTERPRET NTA. Columns that are not identified as a blank or sample are determined to be “pass through” columns that are store in a temporary variable and appended to the data after processing.

Feature Removal and Flagging

If selected by the user, the void volume filter is then applied. This results in any chemical features with a “Retention_Time” value lower than the user-input value to be removed from the dataframe.

Duplicate chemical features are identified by performing matrix comparisons between chemical features’ masses and retention times. A duplicate is identified as a chemical feature whose mass and retention time are both below the user-selected threshold values with another chemical feature – a new column, “Duplicate Feature?”, will receive a value of “1” for these chemical feature pairs.

If any chemical adducts (and/or neutral losses and solvent modifiers) are selected by the user, these species are searched against the data. Each adduct has an associated mass, and similar to the duplicate flagging function, matrix math is applied to identify chemical features that are within the retention time window and mass error (specified by the user) to be (or have) an adduct – this information is recorded in three new columns: “Is Adduct or Loss?”, “Has Adduct or Loss?”, and “Adduct or Loss Info”. The first two columns, “Is Adduct or Loss?” and “Has Adduct or Loss?”, are either 0 or 1 for a given chemical feature based on whether any other chemical features are identified as possible adducts by the matrix math. For chemical features that are (or have) adducts or losses identified, more information is provided in the “Adduct or Loss Info”. The “Feature ID” of the related chemical feature is provided, as well as the adduct or loss species that the two chemical features are suspected to be related by.

The user-submitted tracer file is then merged with the data using rounded mass and retention time columns to generate a list of candidate chemical features that may be a tracer chemical. The observed masses and retention times are then compared, and the user-determined threshold values are used to determine matches between the chemical features and tracers. This information is stored in the new column, “Tracer Chemical Match?”, and appended to the chemical features data. Meanwhile, the statistics that were calculated on the chemical features are appended to the tracers file if the given tracer chemical was identified in the data. This updated tracer file is added as the “Tracer Detection Statistics” sheet to the “QAQC” excel file.

Feature Filtering and Cleaning

Next, statistics (mean, median, standard deviation, coefficient of variation [CV], detection count, and detection percentage) are calculated on sample groups for chemical features (e.g., “Mean Sample_1”, “Median Sample_1”, etc.). Statistics columns and adduct columns (see below) are appended to the data and are recorded in tabular outputs as the “All Detection Statistics” sheet(s) in the “QAQC” excel file.

After statistics have been generated, the data goes through “feature cleaning”, which involves the application of replicate %, CV, and MRL thresholds (all selected by the user) to the chemical occurrences and features. A new, empty (i.e., blank cells) copy of the data is made and used to record passes and fails as the QA/QC thresholds are applied. Each threshold is enacted as a Python mask – a boolean structure that stores a True or False value in each cell according to the logic statement – and the masks are used to populate the empty data copy with strings (“R”, “CV”, and “MRL”) that designate whether a cell failed a given QA/QC filter.

After all filters are applied, cells that passed all filters have their values re-populated. The last step is to determine if, after QA/QC filtering, there are any valid numerical observations of a chemical feature remaining. If so, the chemical feature is left alone. If not, the chemical feature is documented in a new column of the data copy, “Feature Removed?”, as having been removed (the strings from failed occurrences are aggregated here; i.e., “CV, MRL”) and the feature is filtered out of the original data. The annotated data copy is stored as the “Decision Documentation” sheet of the “QAQC” excel file.

At the end of the feature cleaning, the remaining features have their mean value observed in the blank subtracted from the mean value observed in each of the samples. This “BlankSub Mean” replaces the “Mean” columns for each of the sample groups for a given chemical feature, and better represents the actual value of the feature attributable to the study sample. If both positive mode and negative mode data were uploaded by the user, the filtered data are then combined into a single dataframe, and stored as the “Final Occurrence Matrix” sheet of the “QAQC” excel file.

Chemical Results Retrieval

If the user has opted to perform database retrival from DSSTox, a chemical features list is generated from the “Final Occurrence Matrix” sheet. Using an API, the masses of those chemicals are searched against the DSSTox database, returning database information for chemicals that are within a given mass tolerance (default value is +/- 5ppm). This results in a list of candidate chemicals for each feature from the input data that passed the QA/QC filtering. Optionally, hazard information can be appended to the candidate chemicals information by enabling the “Search Cheminformatics Hazard Module” toggle. This data is stored as the “Chemical Results” sheet of the “chemical_results” excel file.

Visualizations

Four visualizations are available under “Load Results” and include the CV scatter plot, decision trees, occurrence heatmap, and run sequence plots, each described in more detail below.

CV Scatter Plot
The CV scatter plot contains two subplots representing the mean abundance versus the measured CV of each sample occurrence. Information on the left subplot shows occurrences in samples that were identified as the “Blanks” in the input files, while the right subplot shows occurrences in all other samples. In each plot, the x-axis is dynamically adjusted to reflect the range of mean abundances across all samples, while the y-axis is fixed to a CV range of 0 – 2.5. In each subplot, a red-dashed line identifies the CV threshold set by the user when inputting parameters; bolded text above the line shows the exact CV value used. Red dots identify sample occurrences linked to tracers (from the tracer input file); all other occurrences are grey. The CV scatter plot may be accessed by clicking on “Load Results” and selecting the “CV Scatter Plot” option. This static visualization is also included in the .zip file downloaded from INTERPRET NTA. An example CV scatter plot is shown in Figure 3.

The CV scatter plot can be used to assess the trends and relationship of CVs observed for tracers, compared to all other features, and to set user expectations for reasonable CV values produced by their analytical methods and data processing tools. Typically, tracers are spiked at concentrations that are known to be well-above any measured or estimated instrumental and method detection limits. Accordingly, tracers are expected to have mean abundances that reflect this spiking level (relative to any measured or estimated dynamic range for an instrument and method). Since the tracers are spiked, CVs should be low if the instrument and method are operationally stable. Under suitable spiking concentrations, many NTA methods can produce CV levels close to that of targeted methods, and often below 0.5. For context, many targeted analysis methods allow CV tolerances of +/- 0.3. As NTA methods are rarely validated in the same way as targeted methods, there is limited guidance for setting an allowable CV threshold for any given NTA method.

Decision Tree
The decision tree generated by INTERPRET NTA shows the pass, fail, and non-detect flags for each sample occurrence across all features. The primary function of the decision tree is to compare how different filter threshold choices change the total number of features and occurrences that are kept or removed at each processing step. The decision tree is split into two sections, called A and B. Both sections have an interface for adjusting the sample and blank replicate thresholds, the CV threshold, and the MRL standard deviation multiplier. When these values are changed the respective decision tree is updated in real time, allowing users to immediately determine the impact of individual filtering thresholds on their input data. The interfaces are separated by a table summarizes the total counts for both features and occurrences that are input to and filtered at each step of the MS1 processing workflow. The interactive decision tree may be accessed by clicking on “Load Results” and selecting the “Decision Tree” option. An example decision tree is shown in Figure 4 below.

Within each decision tree the upper-most level shows the number of unfiltered occurrences or features with at least one replicate detection (Present) and the number for which there were no replicate detections (Missing). The second tree level then shows the number of Present occurrences or features that passed the replicate threshold (Kept) and the number that did not (Removed). The third level shows the number of Kept occurrences or features that had CV values at or below the designed threshold. Any occurrences or features not meeting the threshold are labelled with a CV Flag. Finally, the fourth tree level shows the number of occurrences or features with mean abundance levels at or above the MRL threshold. Any occurrences not meeting the threshold are labelled with a MRL Flag. White boxes and arrows in the tree indicate occurrences that passed QA/QC criteria. All gray boxes and arrows indicate occurrences that were not reproducibly observed or not observed above the MRL. Finally, all red boxes and arrows indicate occurrences that met all QA/QC criteria except the CV threshold. Static images of these trees may be downloaded as .png using the buttons on the left of the plots for “A” and “B” to download the A or B section, respectively.

Decision trees should be used to help optimize thresholds and values used for each filtering step in INTERPRET NTA alongside other tabular and visual outputs generated.

Occurrence Heatmap
The INTERPRET-NTA Occurrence Heatmap shows the pass/fail/non-detect flags for each sample occurrence. The occurrence heatmap is generated using data from the output excel file from INTERPRET-NTA. The interactive occurrence heatmap may be accessed by clicking on “Load Results” and selecting the “Occurrence Heatmap” option. A static version of this visualization is also included in the .zip file downloaded from INTERPRET NTA. An example occurrence heatmap is shown in Figure 5 below.

The occurrence heatmap is a representation of the processing outcomes for all features across study samples. The plot is oriented with features along the bottom axis and samples along the side axis. Within the plot each cell represents a sample occurrence, with the color of the cell denoting the data quality decision for that occurrence. Cells shaded gray are non-detects, defined as those for which the sample occurrence had no peak area in the input file OR those that were found to be below the calculated MRL. Cells shaded red are occurrences where the feature was reproducibly detected above the MRL, but with a CV value greater than the threshold set by the user. Cells in white are those that were reproducibly detected, above the MRL, with a CV value at or below the threshold set by the user.

The occurrence heatmap can be used to assess the effect of the CV threshold set by the user on the entire data set and to compare the number of detected/flagged/measured occurrences across each sample type. For example:

• The sample identified as the “blank” for processing should show no detections (completely gray row).
• Other types of blank samples should also show few or no detections (mostly gray row).
• Samples that are in neat solvent (i.e., matrix-free) should likely show some detections (especially if spiked with native compounds).
• Samples that are in matrix will likely show the most detections and have some amount of “flagged” detections where the CV threshold was exceeded.

The occurrence heatmap can be used alongside other outputs to help decide if your CV threshold should be adjusted.

Run Sequence Plot
The run sequence plots generated by INTERPRET NTA are scatter plots which show the relative abundance of user-input tracer chemicals across a sample sequence. Run sequence plots are generated using data from (1) the input detection file(s), (2) run sequence file(s), and (3) tracer file. Interactive run sequence plots may be accessed by clicking on “Load Results” and selecting the “Run Sequence Plot” option. Once loaded, the run sequence plot(s) should populate as shown in Figure 6 below. A static version of this visualization is also included in the .zip file downloaded from INTERPRET NTA.

The run sequence plots allow you to investigate the performance of individual tracer chemicals to one another and across individual samples. For each plot the y-axis represents the abundance of that tracer across samples, with the range of each plot dynamically set based on the observed abundance. The x-axis for each plot represents the order sample acquisition from the run or analytical sequence. Points on each point are colored based on their user-defined group from the input run sequence file (e.g., Blank, Sample, Spiked).

The tracer run sequence plots should be used to assess the trends and relationship of tracers in the study. Of specific interest is examining batch or run sequence effects. For example:

• If the analytical sequence includes batches of samples run on separate days, the user may observe blocks of samples where tracer response is higher/lower than others.
• If the analytical sequence includes many samples, or samples with especially difficult or “dirty” matrices, the user may observe a decrease in the response across the analytical sequence.

Run sequence plots also allow general examination of total detection frequency across samples. Sub-groupings (e.g., Blank, Sample, Spiked) shown in the plots may also highlight matrix effects such ion suppression (all samples with matrix have lower response than samples in neat solution) or ion enhancement (all samples with matrix have higher response than samples in neat solution). Realistically, these plots often display a combination of run sequence and matrix effects that may require detailed consideration or examination.

Excel QA/QC File

The output excel file from INTERPRET NTA contains all information used to make the plots and reports to the user which features passed QA/QC processing, as well as any annotations identified during processing (Table 6). The data contained on the Final Occurrence Matrix sheet is used as the final data frame of features (defined by a mass and retention time), and responses of those features in samples (as blank-subtracted means).

Table 6.Description of sheets and their contents from the INTERPRET NTA Excel output.

Sheet name	Description of contents	Additional comments
Analysis Parameters	Displays the user-selected or defined parameters used to run INTERPRET NTA. This includes the names of input files and the exact values of thresholds used for future refence
All Detection Statistics	Displays the original input detection matrix list with features (as rows) that were present in the input file and all the original columns included in the original input file (mass, RT, peak area in individual samples, any pass-through columns). It is important to note that this sheet does not reduce/replace any input feature or sample-level information (e.g. which features are present and their peak areas in samples) Additional columns have been added to this sheet that report: 1) Max CV across the feature, 2) Calculated MRL (using the selected multiplier, and at each of 3x, 5x, and 10x multipliers), 3) Whether the feature was identified as a duplicate, adduct, or neutral loss of another feature, and 4) Individual sample replicate stats like mean, median, standard deviation, CV, and detection counts and frequency.	Appended with the ionization mode (Neg) or (Pos)
Tracer Detection Statistics	Displays any tracers that were identified in the data based on the input tracer file given by the user. This sheet contains information about the observed feature (mass, retention time) as well as the difference of the identified feature to the values given in the input file (mass difference, RT difference). This sheet also reports the response of each identified feature across samples.	Peak areas reported on this sheet are completely unmodified and match what would be found in the input detection matrix.Any tracers that were in the input tracer file that did not have a matching feature identified are not included in this sheet.
Tracer Summary	Displays summary information about the tracers identified on the previous sheet. Includes information about mass difference, RT difference, detection frequency, and maximum CV across samples.	This worksheet, at a minimum, should be considered for inclusion in any study QA/QC reports.
Decision Documentation	Displays all features present in the original input detection matrix along and calculations of the outcomes of both features and sample occurences. Additional columns are present to flag when any sample occurrences, or entire features, are dopped or flagged based on INTERPRET NTA processing. Also displays occurrence counts as total possible, total present in the input detection matrix, and the final counts with and without flags.	Individual sample occurrences also display flags to identify which were removed or flagged, and for what reason. When a numerical value is displayed, it is the blank-subtracted mean peak area found for that sample. A numerical value indicates that the occurrence passed each QA/QC threshold.
Final Occurrence Matrix	Displays only the features and the sample occurrences that passed INTERPRET NTA filtering criteria. Includes columns for feature-level information (feature ID, mass, retention time) and INTERPRET NTA identifications (tracer match, duplicate, adduct or neutral loss), and INTERPRET NTA calculations (occurrence counts and percentages). All other columns are for displaying peak areas in samples as “Blank_sub_Mean” values. These peak areas are representative of the steps taken during INTERPRET NTA processing (replicate & CV filtering, MRL checks for non-detects, and blank subtraction)	Samples identified as the “Blanks” for MRL calculations and blank subtraction in your input file are shown here as their raw, unmodified peak areas
Final Occurrence Matrix (flags)	Shares the same structure as Final Occurrence Matrix but includes numeric values for flagged sample occurrences that passed the replication and MRL thresholds, but failed the CV threshold	The user should use flagged values with caution in all downstream analyses
Chemical results	Displays the annotations of features that have a precursor mass (or formula) match to compound(s) in the DSSTox database	Not described in detail here

Report Card

Whitehead et al. (2025) introduces a Report Card, used to quickly summarize the most pertinent QA/QC metrics and outputs from INTERPRET NTA processing. The Report Card summarizes outputs from both visual (CV Scatter Plots, Run Sequence Plots, and Decision Trees) and tabular outputs (Tracer Summary and Tracer Detection Statistics) generated by INTERPRET NTA. The Visualizations that should be included are described in more detail in the section above. Also included are summarized false positive and false negative rates determined from the Confusion Matrices described below. While not an automatically generated output from INTERPRET NTA, these confusion matrices can assist users in optimizing parameters to minimize false negatives and false positives in NTA data. See the Confusion Matrices section below for more information. This Report Card should be included as a table in the Supporting Information of a manuscript or in appendices of reports and can be summarized in the main text.

Table 8.Example Report Card that can be used for INTERPRET NTA reporting.

INTERPRET NTA Output	Summary Quality Metrics
Isotopically-labeled Compound Summary Tables (1)	|mass error|=0.1-6.4 ppm, |RT error|=0.0-0.49 min, max CV=0.33-1.22
CV Scatter Plots (2)	1.2% of unfiltered isotopically-labeled compound occurrences above CV=0.8 threshold
Run Sequence Plots (3)	Minor decrease in abundance across sequence; specific sample outliers
Occurrence Decision Tree (4)	48% of unfiltered occurrences removed
Feature Decision Tree (4)	11% of unfiltered features removed
Confusion Matrices (5)	TPR=97-100%; TNR=38-100%; FNR=0-5%; FPR=0-62%
Note:
1Values from Tracer Detection Statistics and Tracer Summary sheets 2 Values calculated from CV Scatter Plots of Tracer Chemicals 3 Qualitative interpretation from Run Sequence Plots 4 From Decision Trees at Feature and Occurrence level 5 Values from Confusion Matrices.

Confusion Matrices

The use of confusion matrices to calculate false positive (FPR) and false negative rates (FNR) to support quality review of tracer chemical data from INTERPRET NTA was first introduced in Sobus et al. (2025) and subsequently used in Whitehead et al. (2025). While not a generated output of INTERPRET NTA results, the investigation of FPR and FNR can aide in identifying individual tracer chemicals with quality issues and in identifying trends across many tracer chemicals. An example matrix showing the categorization of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) is shown in Table 9. The calculation of FPR, FNR, true positive rate (TPR), and true negative rate (TNR) using these values is shown in the equations below. Results for FPR and FNR can assist the user in deciding if changes to INTERPRET NTA input parameters are necessary, as in Sobus et al. (2025) where use of confusion matrices supported the increase of the MRL multiplier and decrease in the mass tolerance of tracer chemical searches.

Table 9.Example confusion matrix showing the structure of false positive, false negatives, true positives, and true negatives with respect to tracer chemicals.

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	TP	FP	TP+FP
	No	FN	TN	TN+FN
	Sum	TP+FN	TN+FP	TP+FN+TN+FP

True Positive Rate (TPR) = (TP/[TP+FN]) × 100
False Negative Rate (FNR) = (FN/[TP+FN]) × 100
True Negative Rate (TNR) = (TN/[TN+FP]) × 100
False Positive Rate (FPR) = (FP/[TN+FP]) × 100

The presence of tracer chemicals in both the input data (detection matrix) and the output data (final occurrence matrix) from INTERPRET NTA can be examined at both the feature and occurrence level to calculate FPR and FNR. Specifically, confusion matrices examining FPR and FRN of tracer chemicals were applied to (1) unfiltered features in detection matrices, (2) detections in detection matrices, (3) filtered features in final occurrence matrix, and (4) filtered occurrences in final occurrence matrix. Instruction on which data, and where to find the data in either the input or output data, is given in Appendix A and is given for both tracer chemicals from internal standards (those spiked in all samples) and from native compounds (those spiked in only some samples). Appendix B shows the use of these confusion matrices from Whitehead et al. (2025) for both internal standards and native compounds. Results of FPs, FNs, and their corresponding rates should be reported alongside other quality statistics and metrics for NTA data processed with INTERPRET NTA.

Unfiltered Features in Detection Matrices

1. Examine total number of tracer present in Tracer Summary – equal to A
2. Examine total number of tracers not present in Tracer Summary – equal to C
3. Examine the number of features incorrectly assigned to tracers in Tracer Summary – equal to B
4. Determine the number of features not expected as tracers by counting the total number of features present in the input detection matrix minus A – equal to D

Detections in Detection Matrices

1. If B is equal to zero in “Unfiltered Features in Detection Matrices” sum the total number of detections from the column “Total Detection Count” on the “Tracer Summary” sheet – equal to A
   • 1a) If B is greater than zero in “Unfiltered Features in Detection Matrices” sum the total number of detections from the column “Total Detection Count” on the “Tracer Summary” sheet for only the tracers counted in A in “Unfiltered Features in Detection Matrices”– equal to A
     
2. Add the values determined in 2a and 2b – equal to C
   • 2a) Determine the total number of missing detections by multiplying the value of A in the “Unfiltered Features in Detection Matrices” by the number of injections. Subtract the value of A from this value.
   • 2b) Multiply the value of C in the “Unfiltered Features in Detection Matrices” by the number of expected detections (number of injections where the tracer was expected)
3. If B is greater than zero in “Unfiltered Features in Detection Matrices” sum the total number of detections from the column “Total Detection Count” on the “Tracer Summary” sheet for only the tracers counted in B in “Unfiltered Features in Detection Matrices” – equal to B
   
4. Count the total number of detections in the input detection matrix for all features not in counted in A in “Unfiltered Features in Detection Matrices”– equal to D

Filtered Features in Final Occurrence Matrix

1. Set A and C equal to NA
2. Count the number of isotopically-labeled tracers that are present as features in the “Final Occurrence Matrix” sheet given as “1” in the “Tracer Chemical Match” column – equal to B
3. Subtract the value B from the value of A in the “Unfiltered Features in Detection Matrices” – equal to D

Filtered Occurrences in Final Occurrence Matrix

1. Set A and C equal to NA
2. Count the number of occurrences across all samples of labeled tracers that are present as features in the “Final Occurrence Matrix” sheet given as “1” in the “Tracer Chemical Match” column – equal to B
3. Subtract the value B from the value of A in the “Detections in Detection Matrices” – equal to D

Unfiltered Features in Detection Matrices

1. Examine total number of tracers present in Tracer Summary – equal to A
2. Examine total number of tracers not present in Tracer Summary – equal to C
3. Examine the number of features incorrectly assigned to tracers in Tracer Summary – equal to B
4. Determine the number of features not expected as tracers by counting the total number of features present in the input detection matrix minus A – equal to D

Detections in Detection Matrices

1. If B is equal to zero in “Unfiltered Features in Detection Matrices” sum the total number of detections in the injections where native compounds were spiked at sufficient levels– equal to A
   • 1a) If B is greater than zero in “Unfiltered Features in Detection Matrices” sum the total number of detections in the injections where native compounds were spiked at sufficient levels for only the tracers counted in A in “Unfiltered Features in Detection Matrices”– equal to A
2. Subtract the value of A from the value determined in 2a below- equal to C
   • 2a) Determine the maximal possible detections of native compounds in injections where they were spiked at sufficient levels using AC from “Unfiltered Features in Detection Matrices” multiplied by the number of injections where native compounds were spiked at sufficient levels
3. If B is greater than zero in “Unfiltered Features in Detection Matrices” sum the total number of detections in the injections where native compounds were spiked at sufficient levels for only the features counted in B in “Unfiltered Features in Detection Matrices” – equal to B
4. Count the number of detections in the input detection matrix for all features included in D from the “Unfiltered Features in Detection Matrices” features for the injections where native compounds where spiked at sufficient levels– equal to D

Filtered Features in Final Occurrence Matrix

1. Set B and D equal to NA
2. Count the number of native compounds that are present as features in the “Final Occurrence Matrix” sheet given as “1” in the “Tracer Chemical Match” column – equal to A
3. Subtract the value A from the value of A in the “Unfiltered Features in Detection Matrices” – equal to C

Filtered Occurrences in Final Occurrence Matrix

1. Set B and D equal to NA
2. Count the number of occurrences across all samples where native compounds were spiked at sufficient levels that are present as features in the “Final Occurrence Matrix” sheet given as “1” in the “Tracer Chemical Match” column – equal to A
3. Subtract the value of A from the value determined in 3a below- *equal to C**
   • 3a) Determine the maximal possible detections of native compounds in samples where they were spiked at sufficient levels using AC from “Unfiltered Features in Detection Matrices” multiplied by the number of samples where native compounds were spiked at sufficient levels

Unfiltered Features in Detection Matrices

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	29	0	29
	No	1	6,765	6,766
	Sum	30	6,765	6,795

TPR = (29/30) × 100 = 96.67%
FNR = (1/30) × 100 = 3.33%
TNR = (6,765/6,765) × 100 = 100%
FPR = (0/6,765) × 100 = 0%

Detections in Detection Matrices

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	3,560	0	3,560
	No	191	330,769	330,960
	Sum	3,751	330,769	334,520

TPR = (3,560/3,751) × 100 = 94.91%
FNR = (191/3,751) × 100 = 5.09%
TNR = (330,769/330,769) × 100 = 100%
FPR = (0/330,769) × 100 = 0%

Filtered Features in Final Occurrence Matrix

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	NA	18	18
	No	NA	11	11
	Sum	NA	29	29

TPR = NA
FNR = NA
TNR = (11/29) × 100 = 37.93%
FPR = (18/29) × 100 = 62.07%

Filtered Occurrences in Final Occurrence Matrix

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	NA	58	58
	No	NA	3,502	3,502
	Sum	NA	3,560	3,560

TPR = NA
FNR = NA
TNR = (3,502/3,560) × 100 = 98.37%
FPR = (58/3,560) × 100 = 1.63%

Unfiltered Features in Detection Matrices

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	40	0	40
	No	0	6,754	6,754
	Sum	40	6,754	6,794

TPR = (40/40) × 100 = 100%
FNR = (1/30) × 100 = 0%
TNR = (6,754/6,754) × 100 = 100%
FPR = (0/6,754) × 100 = 0%

Detections in Detection Matrices

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	1,040	0	1,040
	No	0	79,409	79,409
	Sum	1,040	79,409	80,449

TPR = (1,040/1,040) × 100 = 100%
FNR = (0/1,040) × 100 = 0%
TNR = (79,409/79,409) × 100 = 100%
FPR = (0/79,409) × 100 = 0%

Filtered Features in Final Occurrence Matrix

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	40	NA	40
	No	0	NA	0
	Sum	40	NA	40

TPR = (40/40) x 100 = 100%
FNR = (0/40) x 100 =100%
TNR = NA
FPR = NA

Filtered Occurrences in Final Occurrence Matrix

		Expected to Map to Tracer Chemicals
_	_	Yes	No	Sum
Mapped to Tracer Chemicals	Yes	239	NA	239
	No	1	NA	1
	Sum	240	NA	240

TPR = (239/240) x 100 =99.58%
FNR = (1/240) x 100 = 0.42%
TNR = NA
FPR = NA